Automated evaluations are provided for every MT model. Click on a model name or the ellipsis in the column to view them.
Phrase Custom AI offers rich data and advanced visual support designed to provide a deeper understanding of custom NextMT model quality:
-
The tab provides a summary of the evaluation results, featuring intuitive visualizations and metadata about the MT model.
-
The table compares the performance of generic versus custom NextMT models across four MT quality metrics. The table has two main sections:
-
Shows automated MT quality scores for Phrase NextMT and a custom NextMT model without TM leverage.
-
Shows automated MT quality scores where TM fuzzy matches are leveraged to adapt MT output.
The column highlights the highest-performing model for each metric.
-
-
The panel provides essential information about the evaluated custom NextMT model.
-
-
The tab provides a graphical representation of MT evaluation results through donut charts, offering a breakdown of evaluated translation segments by quality category.
-
Select the desired MT quality metric from the dropdown menu at the top to benchmark the custom NextMT model against the generic Phrase NextMT model.
-
Hover over each category of the donut chart(s) to view the percentage and number of affected segments for that category.
-
-
The tab presents a segment sample preview from the evaluation set, displaying a list of source segments with relevant baseline and RAG performance scores.
When a segment is selected, the right panel displays:
-
Segment-specific scores and quality level indicators for baseline and RAG performance.
-
A comparison of the translation output generated by custom and generic NextMT models against the reference translation from the dataset. Select to highlight differences against the reference translation.
-